Académique Documents
Professionnel Documents
Culture Documents
Division of Structural Engineering and Structural Mechanics, Department of Civil Engineering, University of California,
Berkeley, California, U.S.A.
SUMMARY
A simple, efficient and versatile finite element is introduced for shell applications. The element is
developed based on a degeneration concept, in which the displacements and rotations of the shell
mid-surface are independent variables. Bilinear functions are employed in conjunction with a reduced
integration for the transverse shear energy. Several examples are tested to demonstrate the effectiveness
and versatility of the element. The numerical results indicate that the shell element performs accurately
for both thick and thin shell situations.
INTRODUCTION
The importance of shell structures and their generic analysis complexities has naturally led to a
reliance on the finite element method for the solution to many types of shell problems. This, of
course, requires a finite element representation of the shell behaviour and over the last 20
years many elements have been developed and employed in a multitude of programs. A
history of shell elements is traced in References 1 , 2 and only a brief summary will be included
here. The papers main theme is the development of a new shell element: the (B)ilinear
(D)egenerated (S)hell. The BDS element attempts to rectify problems inherent to most shell
elements: (a) the limited scope of problems which can be solved and, (b) the high formulation
cost of computing the element stiffness. The former restricts most elements to one class of
shells, either thin or thick shells, depending on the parent theory used for developing the
element. The element stiffness formulation may not be a cost factor for most linear problems
but is of paramount importance in non-linear formulations. Thus the quest for the ideal shell
element that is universally applicable and cheap to use was a motivating force in the present
study.
The approach taken in this paper explores an avenue that was successfully exploited by
Hughes el a13 for developing a simple bilinear plate element. The element in Reference 3 was
based on a one-point quadrature for the transverse shear strain energy. The solution, despite
its simplicity, is surprisingly accurate for both thick and thin plates.
* Graduate Student and Research Assistant. Present address: Division of Structural Engineering and Construction,
Asian Institute of Technology, P.O. Box 2754, Bangkok, Thailand.
0029-5981/79/0214-0179$01.00
@ 1979 by John Wiley & Sons, Ltd.
UdV
Principle of
stationary energy
( U i ) T [ J ( B 3 T B'dY']
D
U'
Discretization
U=N''U'
~~
( L U ) T DLU dV
D.LU
3-D STRESSES
S = ( B q T D B'dS'
ELEMENT STIFFNESS
Principle of
stationary energy
(U31 J ( B ' I
U=N'U'
Discretiza tion
( L U ' D L U ) dS
E U d S
D LU
SHELL STRESS-RESULTANTS
1 I
I
Kinematic compatibiliij
(displ.& rotations/slopes)
dV'
B'dV'
J = I J ( B ID, B J ~ I ~ W
S-(Bq'D
ELEMENT STIFFNESS
Principle o j
stationary energy
( U q T [ J ( B q ' D B'dY"lUJ
( U q r ( B q T D B'U'
I -
0-dVY
3-D STRESSES
B' U'
3-D STRAINS
Kinematic compatibihty
U'
NODAL VARIABLES
DECENERATEDSHELL
U=N'UI
Discretization
MID-SURFACE VARIABLES
CLASSICAL SHELL
Kinematic
shell
assumption
General
constitutive law
3-D STRAINS
Kinematic compatibility
3-D CONTINUUM
bo
181
paragraph will describe the highlights as they relate to the main theme of this paper. The
details of previous shell formulations are contained in the references and the reader is referred
to them for a detailed understanding of the development of previous shell elements.
3 - 0 continuum elements
The three dimensional field equations can be processed to form a 3-D continuum element,
Figure 1. This, of course, produces an element that is ignorant of the usual kinematic
constraints, or assumptions, of most shell problems. Popular among these is: the thickness of
the shell is small compared to the other dimensions. Coupled to this is that the shell generally
carries load via bending and membrane action. The 3-D formulation is now at a distinct
disadvantage: several layers of elements or higher order elements are required to portray
bending. Thus economic considerations usually curtail the usefulness of this element. In
addition, a 3-D element is found to fail at a moderate length to thickness aspect ratio due to
displacement locking through the thickness. Wilson4 suggested the use of incompatible displacement modes to improve the basic 3-D thick shell performance. Dovey followed with a
study of both the reduced integration technique and the idea of adding incompatible modes.
These modifications do produce the desired effect, a more flexible shell behaviour hence better
performance. However, the convergence of these modified 3-D elements cannot be guaranteed.
182
W. KANOK-NUKULCHAI
order differential equations, therefore, the elements require only C'-continuous shape
functions.
Many authors'-'' who developed shell elements based on this concept obtained unsatisfactory results when these elements were applied in the thin shell regime. The difficulty can be
traced to the transverse shear energy which is O [ ( L / h ) 2 ]higher than the remaining terms,
where L / h denotes the element length to thickness aspect ratio. Thus as the thickness
approaches zero, the computed shear stiffness completely dominates, and no effect of the
bending stiffness remains with a finite computer word length. As a result, the element produces
an excessively stiff solution which does not reflect the correct bending behaviour. Physically,
this phenomenon is known as a shear-10cking.~
Several techniques have been studied in an attempt to solve the shear-locking problem.
Wempner et all0 introduced a discrete Kirchhof hypothesis which enforced the Kirchhoff
hypothesis only at the mid-side of the elements. The poor performance of this element was
due, presumably, to too many constraints being enforced. The correction employed in
Reference 10 was to reduce the number of constraints. All shear energy was suppressed and
good results were obtained. The element with no shear energy reverts to being good for only
thin shells hence this solution is wanting. In addition,,thetechnique used in Reference 10 leads
to either unsymmetric coefficient matrices to be solved for the nodal displacements or an
increased number of unknowns if Lagrange multiplier methods are used to satisfy the constraints. Zienkiewicz et al" retained the transverse shear energy but used reduced integration
for quadratic and cubic isoparametric serendipity elementsI2 and obtained good results for
some thin and thick shells. In this formulation, the transverse shear strain energy in the thin
shell situation is a penalty function for the Kirchhoff-Love hypothesis and requires under
integration to avoid the locking effect.I3 For certain boundary conditions, however, this
element shows a tendency to lock in thin shell analysis due to the additional constraints
imposed by the boundary conditions.
183
element is being input in the direction normal to the mid-surface at each node, the position at
any point in the element can be uniquely given in terms of nodal co-ordinates and thicknesses
as
4
x(r, s, t ) =
1 N1(r,s){x'
I=1
+$fh1&j3}
(1)
where XIare the co-ordinates of the mid-surface, h' the thickness and i2f3 the normal unit
vector, all at a node I. The interpolation function N' is bilinear, i.e.,
(2)
184
W. KANOK-NUKULCHAI
The first is normal, while the others are tangent to the midsurface at a point (r, s, 0) on the
mid-surface (Figure 2).
The displacement vector at any point (r, s, t ) in the element can be given in the form:
where ur is the nodal displacement vector on the mid-surface, and uf is the relative nodal
displacement vector produced by a normal rotation at the node. The vector uf is to be
expressed explicitly in terms of the rotation vector, ar,about each of the global axes at the
node. Using the shell assumption that straight normals to the reference mid-surface remain
straight after deformation, the displacement vector based on the local z co-ordinates, produced
by the normal rotation a' about local axes, is w p and expressed by
For infinitesimal rotations, the usual transformations from wf to uf and a" to ar,in view of (7)
lead to
(8 1
u:(t) = ;th'@'a'
where
@I=
in which
-ei3
e23
04,
-efi3
:i3]
eii denotes the direction cosine from global to local co-ordinates, i.e.,
eij= . G,J
(9)
(10)
Substituting (8) into (6) yields the expression of the displacement vector at any point in the
shell element in terms of nodal variables:
Stresses-strains
At any point in the shell element, the local strain components of interest are
The symbol ( . ) / z i defines a derivative with respect to local curvilinear co-ordinates. The
derivation of these strain components can be achieved by using the second-order tensor
185
where ( . ),x, denotes a partial derivative with respect to the global Cartesian co-ordinates and
0 =[&I is the transformation matrix defined in (10).
For a linear isotropic elastic material, the local stress components are obtained from the
usual shell constitutive equations which assume a state of plane stress in each lamina. Accordingly
u(r, s, t) = DE(r, s, t)
(14)
where u(r, s, t)= [alla22a12a 1 3 a23]is the local stress vector and
Ii
x
xc2p
0
0
x+2p
D=
8 8
0
0
p;Ks
0
0
j K ]
in which K , is a shear deformation correction (1.2 is used for BDS), p is the shear modulus,
and is the plane-stress reduced Lame constant, i.e., = vE/(l- v2), where E is the modulus
of elasticity and v is Poisson's ratio.
KN =
(B')TDBJ d V
in which B' is the standard strain matrix relating the local strain vector to the nodal variables
6' = [u', Y'(]'
such that
&(r,s,t)=C~'6'
r
(17)
It is convenient to split the stiffness in (16) into two parts: the transverse shear effect and the
bending and membrane effects. This will allow use of an appropriate order of numerical
integration for each part. Accordingly, let
KN = KK+
where
KZ=[ (Bk)TD,B!,,dV
V
(18)
186
W. KANOK-NUKULCHAI
. ) and L,( . ) are gradient operators with respect to local co-ordinates which
I k
and
L,(f)=
Physically, B: is the strain contribution of the inplane displacements at node I, and B: + tB: is
the strain contribution of the rotations at node I, which also includes the curvature effect.
Taking into account the way the local co-ordinates {z} are constructed, it can be shown via
an orthogonality condition that
L,(t)=O
(28)
187
which results in
I
Bzrn=O
where IJI is the determinant of the Jacobian matrix obtained from (1). To be consistent with
the shell assumption of straight normals, IJ(r, s, t)l can be approximated by IJ(r, s, 0)l. Since B1
and B3 are functions of r and s only, the integral in (30) can be analytically integrated with
respect to t which leads to
The numerical integration is then used to evaluate the remaining integrals in (31) and (32). A
two-by-two Gaussian quadrature is employed for the integral in (31) and, as mentioned
before, a one-point Gaussian quadrature for the integral in (32). This will ensure singularity of
the KY which is necessary in order to avoid the shear-locking phenomenon.
Torsional effect
In using a bilinear-element assemblage to approximate curved shells, convergence is spoiled
by a weakly-restrained torsional mode after the mesh reaches some state of refinement. The
reason can be explained as follows: The BDS element employs six degrees-of-freedom per
node, however, no stiffness corresponding to the torsional-rotation degree-of-freedom exists
locally in the formulation. All the resistance to this rotation at each node I comes directly from
the coupling of the a'-rotations of the non-planar elements surrounding node I. When the
finite element mesh is refined, angles of the kinks between these elements are close to 27r and
the coupling effect is reduced. This weak coupling only generates a minute amount of stiffness
for the torsional rotation. Therefore, any slight disturbance in the generalized load corresponding to this degree-of-freedom can amplify the torsional mode to an unrealistic amount,
which affects the global solution.
This problem is common to many shell elements which use six global degrees-of-freedom
per node. In the past, a fictitious torsional spring was added either locally at the element level,
or in some pseudo-normal direction defined at each node. This technique often is found
unsatisfactory, especially for a flexible system in which an unrealistic amount of strain energy
in the spring can be produced by a rigid body motion.
In a degenerated shell, the rotation of the normal and the mid-surface displacement field are
independent. The idea then is to derive an additional constraint between the torsional rotation
of the normal, a;,and the rotation of the mid-surface, 1/2(dw2/dz1 --dwl/d22).
The deviation of the associated rotation from the mid-surface slope, Figure 3(a), is governed
by the transverse shear strain energy, i.e.,
rS= Ksph L[a;(r,
(33)
188
W. KANOK-NUKULCHAI
In fact, the transverse shear stiffness K, of equation (20) can be obtained directly from the
stationary value of rs.Similarly, the deviation of the torsional rotation of the normal from that
of the mid-surface, Figure 3(b), is assumed to have a governing strain energy,
at the Gauss points. A one-point numerical integration should be used in evaluating the
penalty integral in order to avoid an over-constrained situation similar to shear locking.
To derive a torsional stiffness from equation (34) the local variables are expressed in terms
of global nodal variables, S', by bilinear shape functions. This gives equation (34) in the form,
I T
r t = ( S ) K,S
(36)
189
and
Since two penalty functions are included in the thin shell situation, the penalty factors, K,ph
and K&h, should have magnitudes of the same order. The result displayed in Figure 5 indicates
that the converged solution is insensitive to K , , as long as K , is large enough (>0.1) to
sufficiently restrain the troublesome torsional modes. This insensitivity demonstrates that the
addition of a torsional stiffness will not degrade the behaviour of the system after the torsional
effect is deleted.
All numerical examples in this study employ a value of K , = 10. It should be noted that in
example 2 when the real hyperbolic paraboloid shell geometry is exactly represented, an
identical solution is obtained with or without the torsional stiffness.
NUMERICAL EXAMPLES
T o demonstrate the effectiveness and the versatility of the element, several examples of
different shell situations are analysed.
190
W. KANOK-NUKULCHAI
THICKNESS=0.25
DIAPHRAGM
E-3 x 10 ksi
WEIGHT=009k /sq . f t
( 0 ) THIN S H E L L , H z O 25FT
LL
hi
0
W
W
LL
LL
a:
W
k-
z
W
V
z
9
J
LL
W
i/
( b ) THICK
S H E L L , H = 2 5 FT
0 02577
+
a
,
4
7
7
( 4 x 4 MESH1 0 2893
w
V
--____
0025:
0 02478
002432(2~2MESH)
>
002178
0 02
Figure 5. Effect of torsional stiffness to solutions of thin and thick cylindrical shell roofs
191
I-
GRAPH NO
I
2
3
4
5
6
ELEMENT TYPE
D O FIELEM
CLOUGH-JOHNSON'S FLAT
PLATE (4-TRIANGLED) ELM
AHMAO'S PARABOLIC E L M
AHMAD'S PARABOLIC ELM
W/REDUCED SHEAR INTEG
AHMAD'S CUBIC ELEMENT
ZIENKIEWIC2,TAYLOR &TOO
PARABOLIC W / 2 ~ 2INTEG
SAME AS 5, BUT W/EXPLICIT
INTEGRATION THROUGH
THICKNESS
REF
37
40
12
40
60
ie
40
II
40
22
II
Figure 6. Convergence 01 deflection versus degrees-of-freedom, thin cylindrical shell roof problem
-051
00
100
20"
30"
cp
Figure 7. Cylindrical shell roof, vertical displacement at y
= 1/2
40'
192
W. KANOK-NUKULCHAI
0 010
+
LL
0005
I-
Z
W
W
0
_J
'"
D
-0005
EXACT SHALLOW SHELL THEORY (REF 18)
zn
- + - 8 x 8 ELEMENTS
-+-
3
,n
k
---A---
-0010
- 0 015
4 x 4 ELEMENTS
2 x 2 ELEMEN TS
200
30
100
00
40
(a)
'"-
--
Oo
I -
loo
20
30
0 -*
40
4J
0
I 14
112 y/L
Oo
10'
20'
30'
40'
9
5
(a 1
(b)
Figure 9. Stress resultants for 8-by-8 element mesh. A11 stresses are obtained by the interpolation of values at Gauss
points
193
E = 28500 Ib/in2
-4 xI
4 ELEMENTS
--
s
a
1.0-
z
o_
I/4
I12
314
X /A
Figure 11. Deflection at centre line of hyperbolic shell
N
"CC
ncr.
,A
I-
u,v -BILINEAR
Tr
LL
EXACT(REF 19)
..n
I-
z
w
2.46
2.46-
2.0
2x2
4x4
8x8 12x12
MESH SIZE
Figure 12. Clamped hyperbolic shell, convergence of central deflection
194
W. KANOK-NUKULCHAI
A
Figure 13. Long cylindrical thick shell on line support
Arch dam
This example provides a good test of the element in the case of an arbitrary, thick shell with
varying thickness. Figure 15 shows a doubly curved arch dam known to be type 5 . I 5 , l 6 Details
of the geometry can be found in References 15 and 16. In Figure 16, the downstream
displacement is plotted and compared with the solutions from a thin shell element, a 3-D
continuum element, and a higher-order element.'*
It is remarkable to note that accurate results are attainable with a very crude mesh (16
elements) using this low order element, and are comparable to those of the finer mesh (32
elements) using the higher-order element of Reference 12.
Curved box girder
The last example shows a more complex structure of a curved box girder as detailed in Figure
17. The present element is used for the flanges, curved webs and diaphragm of the box girder.
For webs, a 1-by-2 Gaussian quadrature, Figure 18, is employed for the inplane @-energy
term to avoid the poor beam-bending behaviour which is analogous to the transverse shearlocking in the shell situation. The 2-point integration along the web thickness is necessary to
.-.
UNDEFORMED
0.737~
1CT2
I
ANALYTICAL
SOLUTION
0-0
PRESENT ELEMENT
PRESENT ELEMENT
WITHOUT REDUCED
INT EGRAT10N
2-0 PLAN STRAIN
(2-4ELEMENT
O--O
SCALE OF
L-
195
196
W. KANOK-NUKULCHAI
"
20
10
40
30
50
60
DISPLACEMENT, mm
END
DIAPHRAGM
P=201b
TOP FLANGE
II '
I-WEE
I
A+
IV
r'
I(
BOTTOM FLANGE
( a 1 SECTION A-A
LOCATION OF POINT LOAD
END DIAPHRAGM
( b ) TOP VIEW
r
I
IE.Oin----------y
R = 51 O m
-4
( c ) SECTION B-B
70
197
/ p'2
WEB
( P L A T E ELEMENT)
(DEEP BE
ELEMEN
L-
I
FLANGE
(PRESENT
ELEMENT)
(PRESENT
(PRESENT
ELEMENT W/
OUT-OF- PLANE
EFFECT
NEGLECTED)
A L L 2 x 2 OUT-OF-PLANE
EFFECT IGNORED
- 0 20
1I
--C
2--0.15
-0-
Lw
r-EXTERIOR
WEE
-1
-0.10
cl
-1
+
Lz
-0.05
LLI
>
0 V
0"
so
loo
I
15O
20
25'
30'
35'
40'
J
45"
198
W. KANOK-NUKULCHAI
I00
'
m
+HIGHER
m 75 m
-0-
K
I-
*-I 5 0 5
I-
W 25(3
z
Q
I-
LOCATION OF STRESS
Oo
10"
5O
15O
20"
25'
35O
30'
40'
45'
Figure 20. Tangential stresses on outer face of the bottom flange of box girder
-80,
STRESSES AT M I D - P L A N E
-0HIGHER ORDER ELEMENTS (REF 171
PRESENT ELEMENTS
I
SECTION AT 8.0"
(MID-SPAN)
I
include the equivalent effect of the analytical integration in a classical beam. The diaphragm is
treated as a membrane by deleting the bending stiffness,
The vertical deflection and normal stresses in both flanges are plotted in Figures 19-22. The
solutions are compared with both the experiment and the higher-order finite element solutions
reported in Reference 17. The present element proves to perform surprisingly well in all
aspects.
CONCLUSIONS
The attractiveness of the BDS element is attributed to its efficiency, effectiveness and versa;
tility. The simplest element geometry is chosen so that the element can serve as a convenient
basis for unlimited forms of shell geometry. The degeneration concept, coupled with a reduced
integration technique, produces a shell element which performs accurately in both thick and
199
I00
cn
a
6 80
cn
cn
W
rK
60
cp
I-
A
-0-
ON OUTER FACE
EXPERIMENT (REF 171
HIGHER ORDER ELEMENTS (REF 17)
PRESENT ELEMENTS
+Y
SECTION AT e.00
( MID - SPAN )
STRESSES AT MID-PLANE
-f
-18
- 12
-6
12
18
y, INCH
Figure 22. Tangential stresses in bottom flange at mid-span of box girder.
thin shell situations. With slight modification, the element is also capable of analysing other
types of thin structures, such as a curved web of the curved box girder.
ACKNOWLEDGEMENTS
The author would like to thank his advisors, Profs. R. L. Taylor, T. J. R. Hughes and J. L.
Sackman for the invaluable guidance in this research. He is also indebted to Dr. L. Ovenshire
of the National Highway Traffic Safety Administration of the Department of Transportation
for continued support and interest. The support of the University of California Computer
Center is also gratefully acknowledged.
REFERENCES
1. S. F. Pawsey, The analysis of moderately thick to thin shells by the finite element method, SESM Report No.
70-12, Dept. of Civil Engineering, U.C. Berkeley (1970).
2. H. H. Dovey, Extension of three dimensional analysis to shell structures using the finite element idealization,
SESM Report No. 74-2, Dept. of Civil Engineering, U. C. Berkeley (1974).
3. T. J. R. Hughes, R. L. Taylor and W. Kanoknukulchai, A simple and efficient finite element for plate bending,
Int. J. num. Meth. Engng, 11, 1529-1543 (1977).
4. E. L. Wilson, R. L. Taylor, W. P. Doherty and J. Ghaboussi, Incompatible displacement models, in Numerical
and Computer Methods in Structural Mechanics, (Ed. S. J. Fenves et al), Academic Press, New York, 1973.
5. R. W. Clough and R. J. Johnson, A finite element approximation for the analysis of thin shell, Int. J. Solids
Struct. 4,43-60 (1968).
6. S . W. Key and 2. E. Beisinger, The analysis of thin shells with transverse shear strains by the finite element
method, Roc. Conf.Matrix Meth. Struct. Mech, Wright-Patterson A.F.B., Ohio, AFFDL-TR-68-150, 667-710
(1968).
7. B. M. Irons and K. J. Draper, Inadequacy of nodal connections in stiffness solution for plate bending, AIAA f.3,
965 (1965).
200
W. .KANOK-NUKULCHAI
8. R. J. Melosh, A flat triangular shell element stiffness matrix, Proc. Conf. Matrix Meth. Struct. Mech., WrightPatterson A.F.B., Ohio, AFFDL-TR-66-80,503-514
(1965).
9. S . Utku, Stiffness matrices for thin triangular elements of non-zero Gaussian curvature, A I A A J. 5 , 1659-1667
(1967).
10. G. A. Wempner, J. T. Oden and D. A. Kross, Finite-element analysis of thin shell, J. Engng Mech. Div., ASCE,
94, EM6,1273-1293 (1968).
11. 0. C. Zienkiewicz, R. L. Taylor and J. M. Too, Reduced integration technique in general analysis of plates and
shells, Int. J. num. Meth. Engng., 3, 275-290 (1971).
12. S. Ahmad, B. M. Irons and 0. C. Zienkiewicz, Analysis of thick and thin shell structures by curved finite
elements, Int, J. num. Meth. Engng, 2, 419-451 (1970).
13. 0.C. Zienkiewicz and E. Hinton, Reduced integration, function smoothing and non-conformity in finite element
analysis, J. Franklyn Inst. 302, 443-461 (1976).
14. J. J. Connor and C. Brebbia, Stiffness matrix for shallow rectangular shell element, J. Engng. Mech. Div. ASCE,
93, EM5,43-65 (1967).
15. J. Ergataudis, B. M. Irons and 0. C. Zienkiewicz, Three-dimensional analysis of arch dams and their foundations, Proc. Symp. Arch Dams, Inst. Civ. Engng, London, 37-50 (1968).
16. 0. C. Zienkiewicz, C. J. Parekh and I. P. King, Arc dams analyzed by a linear finite element solution program,
Proc. Symp. Arch Dams, Inst. Civ. Engng, London, 19-23 (1968).
17. A. R. M. Fam and C. Turkstra, Model study of horizontally curved box girder, J. Engng Struct. Div., ASCE, 102,
ST5,1097-1108 (1976).
18. A. C. Scordelis and K. S. Lo, Computer analysis of cylindrical shells, A C I J . 61, 539-561 (1964).
19. S. M. K. Chetty and H. Tottenham, An investigation into the bending analysis of hyperbolic paraboloid shells,
Indian Concrete J., 38,248-258 (1964).
20. 0.C. Zienkiewicz, The Finite Element Method, McGraw-Hill, London, 1977.
21. G. R. Cowper, G. M. Lindberg and M. D. Olson, A shallow shell finite element of triangular shape, Int. J. Solids
Struct. 6 , 1133-1156 (1970).
22. R. L. Taylor, private communications.
23. K. Forsberg, An evaluation of finite difference and finite element techniques for analysis of general shells, Symp.
High Speed Computing for Elastic Struct., IUTAM, Mechanics, Liege 1970.